Modelling Processes in Fractalized Hospitals with Multiagent Systems
and Data Analytics
Marc Premm, Martin Riekert, Achim Klein, and Stefan Kirn
Department of Information Systems 2, University of Hohenheim, Stuttgart, Germany
{marc.premm, martin.riekert, achim.klein, stefan.kirn}@uni-hohenheim.de
Keywords: Process Modelling in Hospitals, Fractalized Organizations, Multiagent-based Modelling, Data Analytics
Abstract: Most approaches for modelling processes neglect the high degree of distributed decision making in the hos-
pital domain where processes are coordinated by local authorities. The paradigm of fractal organizations com-
bined with the decentralized characteristics of distributed Artificial Intelligence may help to understand and
model the problem. This paper presents ongoing research and contributes a meta-model for modelling pro-
cesses in hospitals with multiagent systems as fractals of a logistics supply network and incorporates data
analytics methods to identify dependencies between different fractals. The presented approach is evaluated
by analyzing a hospital scenario involving multiple fractals in a patient-centric process.
1 INTRODUCTION
With a constant change towards profit maximization,
hospitals are forced to apply new methods. To cope
with the increasing cost pressure, approaches from in-
dustrial enterprises seem appropriate and, thus, more
and more hospitals start to make use of process-ori-
entation on their internal workflow (Cleven et al.
2014). Applications in hospitals have to face complex
relationships and dependencies of several depart-
ments that follow locally optimized processes. Clini-
cal pathways are a first approach to cope with the in-
terrelations of multiple departments, but neglect the
limited suggestibility of intra-departmental processes
(Vanberkel et al. 2010). In hospitals, lasting organi-
zational structures are established that partly show a
high degree of autonomy on some levels. As hospital
units structure their internal processes without exter-
nal intervention, these local decisions may influence
the inter-department processes and lead to a subopti-
mal efficacy. To address this issue, the decentralized
decision making process of hospitals has to be in-
cluded in process models to form the basis for analyz-
ing dependencies between multiple departments (i.e.
process fractals). A well-suited approach to model
and analyse such systems with distributed decision
making processes comes with the concept of intelli-
gent, cooperative software agents, and multiagent
systems (MAS). We further propose data analytics
methods to identify interdependent process fractals
and predict time-based parameters to improve coop-
eration among these.
The goal of this paper is (i) to develop a meta-
model for modelling interdependent process fractals
that is suitable for scenarios in hospitals and (ii) to
incorporate data analytics methods to identify the
dependencies between multiple process fractals and
to predict execution times as a basis for improved
cooperation and higher efficacy. The suggested meta-
model is based on two major abstractions: (i) logistics
(the right material in the right quantity, at the right
time and the right place) as well as (ii) the paradigm
of fractal companies introduced by Warnecke (1993).
The paper presents ongoing research and is based on
previous work (Premm & Kirn 2015).
The remainder is structured as follows. Section 2
discusses related work on process management in
hospitals and organizational theory. Section 3
develops a logistics-based meta-model to model
fractal processes. Section 4 presents a data analytics
approach for identifying process fractal dependencies
and predicting execution times. Section 5 presents a
scenario-based evaluation. Section 6 concludes.
2 STATE OF THE ART
This section presents the state of the art on (i) process
management in hospitals, (ii) organizational para-
61
Premm M., Riekert, M., Klein A. and Kirn S.
Modelling Processes in Fractalized Hospitals with Multiagent Systems and Data Analytics.
DOI: 10.5220/0005889500610068
In Proceedings of the Fourth International Conference on Telecommunications and Remote Sensing (ICTRS 2015), pages 61-68
ISBN: 978-989-758-152-6
Copyright
c
2015 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
digms that are used in distributed artificial intelli-
gence and may help to structure processes in hospi-
tals, (iii) the concept of fractal organizational units to
better understand hospital processes, and (iv) the sys-
tematics of logistics tasks and organizational fractals.
2.1 Process Management in Hospitals
In the past, counteracting delays has been performed
by adding additional resources. However, a lack of
resources is in many cases not the cause for delays in
hospitals, but the organization of inter-departmental
processes (Haraden & Resar 2004). Decisions are
made on a local basis and the actors are not aware of
the consequences on other departments or parallel
processes. Haraden and Resar (2004) examine this
problem and evaluate processes of several hospitals
in the United States and the United Kingdom. The au-
thors focus on elective surgeries as well as the sur-
rounding units and found that an overall view on hos-
pital processes may increase the resource efficiency
and thus also the financial performance. However, the
paper is restricted to an overview on possible im-
provements and does not suggest how to enforce pro-
cesses involving multiple units considering the de-
centralized character of decision making in hospitals.
Vanberkel et al. (2010) survey similar approaches
that encompass multiple departments in hospitals fo-
cusing clinical pathways, which aim at eliminating
the ambiguity of patient care trajectory. While other
modelling approaches optimize all patient types in
one department, the scope of clinical pathways is one
patient type with all relevant departments. However,
this point of view neglects different types of patients
competing for the same resources. For Vanberkel et
al. (2010), the optimization of clinical pathways is the
first step before quantitatively optimizing the internal
processes of single departments.
The literature also provides work that specifically
addresses the distributed nature of the decision prob-
lem: Murray et al. (Murray et al. 2014) take a patient-
centric perspective and suggest to use software agents
representing relevant actors of patient care trajectory
(e.g. patient, physician, unit). The agent interaction
protocol takes over the coordination of resources as
well as involved actors while coping with the decen-
tralized nature of the processes. The distributed na-
ture of the decision making process as well as the
huge dynamic caused by the mixture of planned and
emergency instances of numerous individual tasks
lead to a high complexity of the process landscape in
hospitals. The organizational units as well as the pro-
cesses in their responsibility can be considered as
fractals of the organization. Agent-based approaches
from Distributed Artificial Intelligence (DAI show
great potential to cope with this complexity.
2.2 Organizational Metaphors in Mul-
tiagent Systems
In the last decades, researches in DAI have developed
a paradigm called MAS, which is suitable for scenar-
ios with multiple actors deciding about their actions
on a local knowledge basis. The main focus of re-
search has been on an increased flexibility facing pre-
viously unknown environmental circumstances.
However, there are also approaches that cope with or-
ganizational stability required in hospital scenarios.
According to DAI/MAS researchers with back-
grounds in management science, “organization” is a
metaphor that can be useful in helping to describe, to
study, and to design distributed software systems
(Malone 1987; Fox 1981). Compared to organiza-
tional theories in management, however, MAS/DAI
still lacks similar fine-grained concepts and instru-
ments for describing, analyzing, thus understanding
and designing organizational phenomena within
agent-based systems.
Approaches from DAI that involve single prob-
lem solving experts can be compared with the per-
spective of management science, in which organiza-
tions are systems that pool individual resources in or-
der to gain additional benefits for all of their mem-
bers. From an organizational perspective this ap-
proach implements the concept of dividing labor
among a set of individuals each possessing a particu-
lar capabilities profile (Gasser 1992). As an immedi-
ate consequence, distributed problem solving leads us
to role concepts – e.g., the role concept of the C-Net
system (Davis & Smith 1983), in which manager and
contractor roles coordinate the execution of tasks.
However, the definition of roles in DAI is quite dif-
ferent to organizational theories in management sci-
ence. The latter refers to a role as a precise definition
of expected behavior of a particular member of an or-
ganization.
Management science considers organizations
mainly from a social science perspective. This per-
spective builds upon the basic assumption that hu-
mans form an enterprise in order to fulfill a concrete
market demand (e.g., production of autonomous
cars). Organizational rules and definitions (e.g., defi-
nition of positions) are required to coordinate the di-
vision of labor, the behavior of employees, and all op-
erational processes to produce, sell, and maintain
goods and services. It is well understood, that enter-
prises need stability with respect to their suppliers and
customers, to their employees, and to their infrastruc-
tural, technical and financial production factors. It is
well understood, too, that the increased dynamics of
their environments (e.g., changing consumer behav-
iors, changing market demand, changing market
structures, changing market coordination, etc.) does
Fourth International Conference on Telecommunications and Remote Sensing
62
also require an increase of organizational flexibility.
Approaches from distributed artificial intelligence
may serve this kind of flexibility, but lack stability in
terms of organizational structures. In the healthcare
domain, however, both principles are necessary to
fulfill patient care. The paradigm of fractal organiza-
tions by Warnecke (1993) enables using these two ap-
proaches simultaneously.
2.3 From Fractal Processes to Fractal
Organizational Units
It has been argued that the enterprise of the future will
be radically decentralized, in order to meet the chal-
lenges of the increasing complexity of its environ-
ment, and the dynamics of world-wide competition.
Decentralization involves the allocation of autonomy,
resources, and responsibilities to deeper levels of the
organizational hierarchy (for instance, see work of
Tapscott & Caston (1993) or Warnecke (1993)). This
requires enterprises to replace hierarchical planning
by more decentralized concepts of coordination like
the MAS paradigm introduced above. In turn, auton-
omous organizational departments need to exhibit im-
proved capabilities in terms of intelligence and self-
reference than they do today. This has given rise to
the notion of organizational fractals (Warnecke
1993). Organizational fractals are characterized by
the following major criteria (Warnecke 1993): (i)
self-organization and self-optimization, (ii) goal ori-
entation, (iii) dynamic, as well as (iv) self-similarity.
The last criterion of self-similarity describes the
structural characteristics of the organization as well
as the modalities of generating added value. The self-
similarity between different fractals enables resource
sharing especially for informational resources and
thus is especially interesting as it enables to build
complex systems on simple and reoccurring modules.
In the case of hospitals, one can think of several lo-
gistic tasks that have to be fulfilled for patient care.
Whereas the patient itself undergoes multiple differ-
ent process steps that show self-similarity in their in-
ternal structure. Findings from logistics may be trans-
ferred to the hospital domain and may serve to im-
prove processes in hospitals with their fractal organi-
zations.
2.4 Systematics of Logistics Tasks and
Organizational Fractals
Logistics aim at supplying a requesting entity with the
right good (quantity and quality), at the right time and
the right place at minimal costs. The spatiotemporal
transformation of goods is the rudimental capability
of logistics systems. The involved processes can be
distinguished into the following categories (Pfohl
2004): (i) Core processes of goods flow (transport,
transshipment and storage processes), (ii) supporting
processes, e.g. packaging processes and (iii) order
transmission and processing processes. A generic ex-
ample from the manufacturing industry would be the
storage of a resource (temporal transformation) that
has to be prepared for pickup (transshipment), trans-
ported to the targeted destination (spatial transfor-
mation), prepared for further processing (transship-
ment), physically adapted (production), again pre-
pared for pickup (transshipment) and so on. This ele-
mentary example shows that the core logistics pro-
cesses occur continually
The widespread visualization as a graph is do-
main-independent and enables also logistics networks
as an extension of a logistics supply chain (Domschke
2008). Dependent on the specific modelling goal,
there are numerous approaches for formalizing logis-
tics tasks. Besides business driven approaches like the
Architecture of Integrated Information Systems
(ARIS), which provides general means for business
process modelling (Scheer & Nüttgens 2000) and the
Supply Chain Operation Reference (SCOR) Model,
which is an industry-independent framework for eval-
uation and improvement of supply chains (Stewart
1997) a huge range of quantitative decision models
exist in literature. Quantitatively parameterized math-
ematical models are mainly used for planning and de-
cision making, but usually involve only a restricted
number of parameters (Scholl 2008). With these
mathematical models numerous variants of supply
chain optimization problem can be addressed. How-
ever, these models generally assume some central de-
signer that is able to enforce a production plan to all
instances of the supply chain. In real-world scenarios
this is usually not the case, especially in hospital sce-
narios in which single departments remain highly au-
tonomous in their internal processes.
The organizational fractals involve a maximum
degree of local autonomy, self-control, and self-or-
ganization skills. Aiming to maximize their local util-
ity (for instance, in terms of profit), organizational
fractals decide on their own whether they are willing
to cooperate, or to collaborate with other organiza-
tional units. There is no direct means by which frac-
tals can be compelled to behave in a certain manner.
The single acceptable way to control the behavior of
an organizational fractal, or of a group of cooperating
fractals, is through designing a globally consistent
system of aims and objectives (Warnecke 1993).
However, due to bounded rationality, organizations
are, in most cases, not able to establish consistent goal
hierarchies. Instead, the different goals that exist
within an organization are more or less inconsistent,
the knowledge about goals and relationships between
them remains necessarily incomplete, uncertain,
fuzzy, and sometimes even wrong.
Modelling Processes in Fractalized Hospitals with Multiagent Systems and Data Analytics
63
Organizational fractals form organizationally sta-
ble parts of an enterprise and have well-defined inter-
faces to their environments. They execute locally
well-defined production functions (transformations),
and they are supposed to guarantee a maximum of in-
ternal stability in terms of, e.g., their operations and
processes, their requests for resources, their availabil-
ity, and their responsiveness. Their flexibility results
from their capability to cooperate, and even merge
with other fractals in order to create a more complex
fractal, if required.
3 MODELLING FRACTALS
WITH MULTIAGENT
SYSTEMS
To address the complex nature of organizing pro-
cesses in hospitals, this section combines the para-
digm of organizational fractals from management sci-
ence with MAS from DAI and proposes a meta-model
for modelling fractals from a supply network perspec-
tive.
3.1 A Fractal Supply Network
Perspective
The transportation of goods and the systematics men-
tioned in section 2.4 are independent of a certain do-
main and the mentioned types of processes show sim-
ilar characteristics: Goods have to be transported,
handled and stored. In general, this is even independ-
ent of the fact, whether the good in question is physi-
cal or informational. For information goods the bor-
der between these core processes and the order trans-
mission or processing might diminish as no physical
good is present. In this case, the core process is an
information flow just like the order processes.
Independent of the physical presence of a good, it
can be observed that supply chains are in many cases
divided into different fractals. These fractal are auton-
omous and cannot be fully controlled from a macro
perspective. Depending on the context, these fractals
might be whole enterprises (e.g. in a manufacturing
supply chain) or different departments (e.g. in a hos-
pital) that show a certain amount of autonomy.
Hence, the overall process cannot be planned in detail
against the motivation of the single fractals.
3.2 Multiagent Systems
With its focus on distributed decision making, the
paradigm of MAS seems well suited for the local au-
thorities in the hospital domain. Since the emergence
of the multiagent paradigm numerous MAS have
been developed for various domains, e.g. manufactur-
ing and logistics, and in most cases the design is fo-
cused on specific issues (Stockheim et al. 2004). Alt-
hough developed independently, the different MAS
cannot be viewed as separated autarkic systems as
they interrelate with each other in many ways. The
organizational structure between two or more inde-
pendently developed MAS usually involves the rela-
tions between the represented real world organiza-
tions. The technical as well as the organizational
question has been addressed by the platform
Agent.Enterprise in a logistic scenario (Woelk et al.
2006). Agent.Enterprise is not restricted to intra-or-
ganizational value chains already represented by
MAS, but integrates multiple instances of these into
inter-organizational supply chains. This combination
of multiple MAS is called a multi-multiagent system
and works cross-organizational. Each MAS remains
locally controlled, but obtains features of inter-organ-
izational communication and cooperation to further
increase flexibility and decrease costs. In Agent.En-
terprise each MAS plans and optimizes its logistic
and production processes individually, but informs
other systems of unforeseen and potentially disturb-
ing events. On the basis of this information exchange,
plans of other MAS may be adapted or inter-organi-
zational contracts may be renegotiated (Woelk et al.
2006).
3.3 Meta-Model
In logistics supply chains one can find different levels
of organizational structure, e.g. in a manufacturing
supply chain, there are usually different companies
that work together for one final good. Thus, we can
distinguish between intra- and inter-organizational
structures, e.g. the intra-organization structure of a
company is embedded into the inter-organizational
structure of the supply chain that involves various
other companies whose behavior is not controllable,
but has to be motivated. Analogously, processes in
hospitals are characterized by highly autonomous de-
partments that can only be limitedly controlled by the
central hospital process management. This leads to
Figure 1: Organizational fractal.
Fourth International Conference on Telecommunications and Remote Sensing
64
fractal processes within the hospital where each de-
partment again can be represented by a single MAS.
Independent of a certain domain, network-wide
processes consist of flexibly coordinated fractals be-
ing under local control of complex agents, e.g. a sin-
gle MAS. Two dependent organizational problems
evolve: (i) the intra-organizational structure of each
MAS that may differ significantly and (ii) the overall
inter-organizational structure that aims at a final prod-
uct and that is not able to fully control the single pro-
cess fractals. Each fractal has a logistics task based on
domain independent types: (i) spatial transformation
in form of a transportation process, (ii) temporal
transformation in form of storage as well as (iii) phys-
ical transformation in form of a production process.
The single fractals are represented by a MAS with in-
terfaces to form a supply chain.
The internal workflow of each process fractal is
only in a small extent influenceable from an external
position. The operational sequences performed by the
involved actors may be affected by incentives, but
cannot be controlled directly. Hence, for modelling
process fractals in logistic processes, it is necessary to
have a modelling language that allows to abstract
from the internal workflow within a process fractal.
Table 1 shows the meta-model for modelling domain
independent logistic process that show characteristics
of fractalization. Figure 1 shows an example of a pro-
cess fractal involving the modelling elements de-
scribed above. The elements are arranged to represent
a process fractal with a number of interacting actors
and two interfaces.
Table 1: Meta-Model.
Label
Symbol
Process
Fractal
Actor
Interface
Interaction
Path
Process
Flow
As described in section 2.4, logistic processes in
many domains show self-similarity and can be re-
duced to three types of processes: (i) storage, (ii)
transshipment, and (iii) transport. From a logistic per-
spective production processes can be interpreted as
storage processes, as the product or service has no in-
fluence on the logistic system for a certain time and,
thus, is transformed in a temporal manner.
3.4 Formalizing Logistics Tasks
The combination of different process fractals is a cen-
tral feature of the proposed meta-model. The combi-
nation of process fractals that are independent from a
decision making perspective allows to form logistic
supply chains. While each fractal only performs sim-
ple tasks the combination of different fractals may
serve to solve tasks with a higher complexity. Error!
Reference source not found. shows an example of a
combination of different process fractals: Between
transport and storage process fractals, usually, a trans-
shipment process fractal has to be involved to achieve
compatibility. In a flow of goods scenario this might
be the forklift that allows for transshipping goods in
a high-bay warehouse to the transporting truck. How-
ever, these process fractals also match for scenarios
in a hospital domain, e.g. the patient has to be reposi-
tioned (transshipped) from the transportation bed to
the surgical table before surgery (see section 5).
These process fractals can be arranged to different
kind of processes. Figure 3 shows three elementary
types: (i) the single-tier system with only two con-
nected fractals are involved, (ii) the multi-tier system
with different interconnected tiers, as well as (iii)
Figure 3: Basic structures of logistics systems.
Figure 2: Combination of process fractals.
Modelling Processes in Fractalized Hospitals with Multiagent Systems and Data Analytics
65
combined systems that also have connections be-
tween non-consecutive tiers. Processes may also split
up at break-bulk points and may be joint at consoli-
dation points.
4 DATA ANALYTICS FOR
FRACTALIZED PROCESSES
For modelling and better coordinating and supporting
logistic processes among fractals with MAS, methods
for data analytics can be used to (1) identify fractals
in the first place, and (2) predict parameters of the
fractals such as the start, duration, and end of individ-
ual logistic tasks of different types. The identified
fractals can be used for modelling the logistics of an
organization with MAS. The predicted parameters
can be used by MAS that represent fractals to support
and improve coordination among fractals by better
anticipating logistic tasks. The most important pre-
requisite for applying data analytics to the described
respect is the availability of large amounts of data that
allows describing and predicting the fractals’ param-
eters. Nowadays, this seems less of a problem as more
and more data emerges and becomes available due to
new types of sensor systems and information systems
used in the scope of logistics, e.g. electronic
healthcare records and advanced medical devices
(Manyika et al. 2011).
For identifying fractals from data, the traces of the
logistic tasks within an organization have to be col-
lected and made available for analysis. The data
should comprise time stamps and locations of each
individual logistic task (i.e., events of starting and
completing a logistic task) and a unique reference to
the logistic goods across an organization. By sorting
the tasks by the time stamps of starting and complet-
ing events, the routing of goods can be identified and
the duration of tasks can be measured. Aggregating
the (most frequent) routes in a graph-based model can
help to identify the most important routes and also
waiting bottlenecks across fractals can be identified.
For conducting this type of data analytics, several
software tools are available. For instance, the tool
proM can be used (Van Der Aalst et al. 2009).
Nowadays, predicting logistic tasks in an organi-
zation is often accomplished by human estimates.
Theses are often too coarse-grained and the resulting
imprecision leads to bad coordination among fractals
and frustration in the implementation of fractalized
logistic tasks. Methods for data analytics can be used
to more effectively predict all three types of logistic
tasks of process fractals.
Predictive data analytics is to create a prediction
model in a data-driven way, which maps several pre-
dictive variables to the variable to be predicted (here:
parameters of logistic tasks). Finding the optimal
mapping can be well accomplished by machine learn-
ing methods. Machine learning is the ability to im-
prove performance on a task with increasing experi-
ence (Mitchell 1997). Performance is measured in
terms of the error of the prediction model’s output vs.
actual outcomes as described in a historic dataset. In
the last decade the performance of Machine learning
has strongly increased due to the availability of suffi-
cient training data, computational resources and the-
oretical improvements (Vapnik 2000).
Figure 4 outlines the principle approach of ma-
chine learning (Vapnik 2000): the explanatory or pre-
dictive input variables created by the generator are
transformed. The vector transformation makes sure
that variables are represented as real numbers. Further
types of transformations are also possible that might
improve the ability of the method to create an accu-
rate prediction model. The input variables are paired
with the variable to predict, which is to be provided
by a supervisor, e.g. a human annotator. These pairs
are used by a so called learning machine to create a
prediction model, which maps the input variables to
the variable to predict 𝑦̂. With the created model, new
data of the input variables can be used to predict the
variable of interest.
For the data-driven creation of prediction models,
the machine learning method Support Vector Regres-
sion (SVR) can be used (Drucker et al. 1997). The
SVR method is a Support Vector Machine (SVM;
Boser et al. 1992) for regression tasks. The input and
output variables are real numbers. But also textual in-
put can be incorporated by means of n-gram based
text representations (Joachims 1998). Textual docu-
ments are transformed into a vector space representa-
tion by means of determining the frequency of occur-
rence of each n-gram of words within a document and
within a whole corpus of documents. Typically, uni-
grams or bigrams are used.
generator supervisor
learning
machine
y
vector
transformation
x
Figure 4: Machine Learning (Vapnik 2000).
Fourth International Conference on Telecommunications and Remote Sensing
66
The following introduction to SVR is based on
Smola & Scholkopf (2004). Given training data
{(
𝑥
1
, 𝑦
1
)
, … ,
(
𝑥
𝑙
, 𝑦
𝑙
)}
⊂ 𝑋 × ℝ, where 𝑋 is the input
space. SVR determines a function f(x) that is as flat
as possible and has at most 𝜀 distance from the actual
target 𝑦
𝑖
. To allow a higher distance than 𝜀 this algo-
rithm is extended by incorporating a cost parameter
(Smola & Scholkopf 2004). For putting in place data
analytics for the fractals of an organization, a respec-
tive data handling and software architecture is re-
quired. The architecture needs to support the desired
analytic tasks. Analytics can be conducted either in an
offline or online fashion. Offline analytics means to
sample large amounts of data, comprising predictive
variables and also the variables to predict. The data is
used to create the prediction model, which is then ap-
plied unchanged on new data. The online approach
would try to continuously improve the model once
new data becomes available.
5 EVALUATION
Outlined below is a scenario of a patient process,
which evaluates the effectiveness of our approach for
the improvement of the cooperation among hospital
process fractals to improve the overall efficacy. Note
that such a process might arise during emergency and
regular operations and therefore follows the patterns
of reoccurring fractals as described in section 2.4.
The process comprises the following steps: (1)
a patient is brought from the ward to the operation
section, (2) the patient is moved to a bed in the sur-
gery section, (3) the patient is transported to the oper-
ating room, (4) the patient is repositioned to a surgical
table, (5) the surgery takes place, (6) the patient is re-
positioned again to a hospital bed and (7) moved to a
postanesthesia recovery.
Figure 5 shows the mapping of the procedural
steps into the fractal constructs. The dashed circle
represents a fractal, i.e. an autonomously-organized
hospital unit. The solid interconnected circles indicate
interchangeable agents of the organization. The solid
boxes between the fractals represent their interfaces.
The procedural steps of the scenario are mapped to
following fractal constructs: (1) transport, (2) trans-
shipment, (3) transport, (4) transshipment, (5) stor-
age, (6) transshipment, and (7) transport.
In this scenario several problems occur if the
prediction of the process time is imprecise. First, all
steps are subsequent and therefore an imprecise pre-
diction of the duration of a step will directly suspend
the earliest initiation of the following steps. Second,
interdependencies of resources like specialized sur-
geons, medical devices and operating rooms further
delay surgeries in this or other operating rooms.
Third, due to the previously named problems, the
planning of the hospital time is difficult due to the
high variance in the actual execution of plans, which
leads to the allocation of fewer resources to planning
and also decreases commitment of the staff to the
plan, which further increases the prediction error.
By means of the fractal based modelling ap-
proach it is possible to understand the limits of pro-
cess planning. One can easily recognize that pro-
cesses may only be planned on a certain level of ab-
straction. On a more detailed level, the process exe-
cution is always performed by a certain set of in-
volved agents and, thus, can only be indirectly influ-
enced. However, these fractals contain dependencies
among each other that have to be recognized to opti-
mize process execution, e.g. the execution time of
process steps within a fractal may also be relevant for
process steps within other fractals.
The usage of data analytics for the prediction of
process times improves the hospital organization by
following aspects. First, due to machine learning the
start and end time of the fractals can be predicted with
low prediction error. Therefore, the planning error is
directly reduced. Second, the confidence for the pre-
diction can be estimated. This information allows
scheduling surgeries with low prediction confidence
in spots that have as few as possible interdependen-
cies with other surgeries. Third, process times can be
predicted up to the minute. These predicted process
times can be communicated to other affected process
fractals without involvement of a human, which al-
lows the automatic updating of process times of emer-
gency and regular surgeries when new information
becomes available.
6 CONCLUSION
This research contributes a meta-model for frac-
talized organizations from a logistics perspective,
which is used for modelling hospital processes. The
Figure 5: Fractal model of hospital scenario.
Modelling Processes in Fractalized Hospitals with Multiagent Systems and Data Analytics
67
proposed meta-model forms the basis for data ana-
lytic methods aiming to identify dependencies be-
tween multiple fractals. The contribution has been
evaluated by a scenario-based evaluation and is
planned to be validated in a field study in future work.
However, first results show great potential for model-
ling hospitals with the paradigm of fractal organiza-
tions. With mostly independently organized units,
hospitals show a high level of fractalization and, thus,
are predestined for modelling processes following the
paradigm of organizational fractals.
Together with data analytics focused on hospital
needs, the dependencies between different fractals
can be identified and parameters of fractals such as
process duration can be predicted for the benefit of
increasing patient throughput as well as to improve
patient care significantly. A detailed investigation
will be subject to further research.
ACKNOWLEDGEMENTS
This work has been supported by the project
InnOPlan, funded by the German Federal Ministry for
Economic Affairs and Energy (BMWi, FKZ
01MD15002).
REFERENCES
Van Der Aalst, W.M.P. et al., 2009. Prom: The process
mining toolkit. In Business Process Management
Demonstration Track.
Boser, B., Guyon, I. & Vapnik, V., 1992. A training
algorithm for optimal margin classifiers. In 5th Annual
ACM Workshop on Computational Learning Theory.
pp. 144–152.
Cleven, A.K. et al., 2014. Process management in
hospitals: an empirically grounded maturity model.
Business Research, 7, pp.191–216.
Davis, R. & Smith, R.G., 1983. Negotiation as a metaphor
for distributed problem solving. Artificial Intelligence,
20(1), pp.63–109.
Domschke, W., 2008. Grundlagen der
Betriebswirtschaftslehre – Eine Einführung aus
entscheidungsorientierter Sicht 4th ed., Springer.
Drucker, H. et al., 1997. Support vector regression
machines. Neural Information Processing Systems, 9,
pp.155–161.
Fox, M.S., 1981. An organizational view of distributed
systems. IEEE Transactions on Systems, Man and
Cybernetics, 11(1), pp.70–80.
Gasser, L., 1992. DAI Approaches to Coordination. In N.
M. Avouris & L. Gasser, eds. Distributed Artificial
Intelligence: Theory and Practice. Kluwer, pp. 31–51.
Haraden, C. & Resar, R., 2004. Patient flow in hospitals:
understanding and controlling it better. Frontiers of
health services management, 20(4), pp.3–15.
Joachims, T., 1998. Text categorization with support
vector machines: Learning with many relevant
features. In 10th European Conference on Machine
Learning. Lecture Notes in Computer Science. pp.
137–142.
Malone, T.W., 1987. Modeling Coordination in
Organizations and Markets. Management Science,
33(10), pp.1317–1332.
Manyika, J. et al., 2011. Big data: The next frontier for
innovation, competition, and productivity,
Mitchell, T., 1997. Machine Learning, McGraw Hill.
Murray, J., Widmer, T. & Kirn, S., 2014. Agent-based
process coordination among hospital units. In
Multikonferenz Wirtschaftsinformatik. pp. 763–774.
Pfohl, H.-C., 2004. Logistiksysteme –
Betriebswirtschaftliche Grundlagen 7th ed., Springer.
Premm, M. & Kirn, S., 2015. From Cooperating Agents to
Cooperating Multiagent Systems: A Survey. In
Proceedings of the 13th German Conference on
Multiagent System Technologies. Springer.
Scheer, A.-W. & Nüttgens, M., 2000. ARIS Architecture
and Reference Models for Business Process
Management. In Business Process Management:
Models, Techniques, and Empirical Studies. pp. 376–
389.
Scholl, A., 2008. Grundlagen der modellgestützten
Planung. In D. Arnold et al., eds. Handbuch Logisitk.
Heidelberg: Springer.
Smola, a J. & Scholkopf, B., 2004. A tutorial on support
vector regression. Statistics and Computing, 14,
pp.199–222.
Stewart, G., 1997. Supply Chain Operations Reference
Model ( SCOR ): the First Framework for Integrated
Supply-Chain Management. Logistics Information
Management, 10(2), pp.62–67.
Stockheim, T. et al., 2004. How to build a multi-multi-
agent system the agent.enterprise approach. In ICEIS
2004 - Proceedings of the Sixth International
Conference on Enterprise Information Systems. pp.
364–371.
Tapscott, D. & Caston, A., 1993. Paradigm Shift: The
New Promise of Information Technology, New York:
McGraw Hill.
Vanberkel, P.T. et al., 2010. A Survey of Health Care
Models that Encompass Multiple Departments.
International Journal of Health Management and
Information, 1.
Vapnik, V.N., 2000. The nature of statistical learning
theory, Springer.
Warnecke, H.-J., 1993. Revolution der
Unternehmenskultur 2nd ed., Heidelberg: Springer.
Woelk, P.-O. et al., 2006. Agent.Enterprise in a Nutshell.
In S. Kirn et al., eds. Multiagent Engineering – Theory
and Applications in Enterprises. Heidelberg: Springer.
Fourth International Conference on Telecommunications and Remote Sensing
68